GNN for simultaneous leak localization in water networks

A Graph Neural Network model localizes a variable number of simultaneous leaks in water distribution networks without predefined leak counts or exhaustive training combinations. In the 31-node Hanoi network, the GNN outperformed random forest classifiers by 20% and MLP models by 12% under sparse sensor coverage.